1. Supercomputing in Plain English Stupid Compiler Tricks
Henry Neeman
Director
OU Supercomputing Center for Education & Research
October

2. Outline Dependency Analysis What is Dependency Analysis?
Control Dependencies
Data Dependencies
Stupid Compiler Tricks
Tricks the Compiler Plays
Tricks You Play With the Compiler
Profiling
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

3. Dependency Analysis

4. What Is Dependency Analysis?
Dependency analysis describes of how different parts of a program affect one another, and how various parts require other parts in order to operate correctly.
A control dependency governs how different sequences of instructions affect each other.
A data dependency governs how different pieces of data affect each other.
Much of this discussion is from references [1] and [5].
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

5. Control Dependencies Dependencies affect parallelization!
Every program has a well-defined flow of control that moves from instruction to instruction to instruction.
This flow can be affected by several kinds of operations:
Loops
Branches (if, select case/switch)
Function/subroutine calls
I/O (typically implemented as calls)
Dependencies affect parallelization!
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

6. Branch Dependency y = 7 IF (x /= 0) THEN y = 1.0 / x END IF
The value of y depends on what the condition (x /= 0) evaluates to:
If the condition evaluates to .TRUE., then y is set to 1.0/x.
Otherwise, y remains 7.
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

7. Loop Carried Dependency
DO i = 2, length
a(i) = a(i-1) + b(i)
END DO
Here, each iteration of the loop depends on the previous iteration. That is, the iteration i=3 depends on iteration i=2, iteration i=4 depends on i=3, iteration i=5 depends on i=4, etc.
This is sometimes called a loop carried dependency.
There is no way to execute iteration i until after iteration i-1 has completed, so this loop can’t be parallelized.
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

8. Why Do We Care?
Loops are the favorite control structures of High Performance Computing, because compilers know how to optimize their performance using instruction-level parallelism: superscalar, pipelining and vectorization can give excellent speedup.
Loop carried dependencies affect whether a loop can be parallelized, and how much.
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

9. Loop or Branch Dependency?
Is this a loop carried dependency or a branch dependency?
DO i = 1, length
IF (x(i) /= 0) THEN
y(i) = 1.0 / x(i)
END IF
END DO
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

10. Call Dependency Example
y = myfunction(7)
z = 22
The flow of the program is interrupted by the call to myfunction, which takes the execution to somewhere else in the program.
It’s similar to a branch dependency.
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

11. I/O Dependency X = a + b PRINT *, x Y = c + d
Typically, I/O is implemented by hidden subroutine calls, so we can think of this as equivalent to a call dependency.
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

12. Reductions Aren’t Dependencies
array_sum = 0
DO i = 1, length
array_sum = array_sum + array(i)
END DO
Other kinds of reductions: product, .AND., .OR., minimum, maximum, index of minimum, index of maximum, number of occurrences of a particular value, etc.
Reductions are so common that hardware and compilers are optimized to handle them.
Also, they aren’t really dependencies, because the order in which the individual operations are performed doesn’t matter.
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

13. Data Dependencies
“A data dependence occurs when an instruction is dependent on data from a previous instruction and therefore cannot be moved before the earlier instruction [or executed in parallel].” [6]
a = x + y + cos(z);
b = a * c;
The value of b depends on the value of a, so these two statements must be executed in order.
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

14. Output Dependencies x = a / b; y = x + 2; x = d – e;
Notice that x is assigned two different values, but only one of them is retained after these statements are done executing. In this context, the final value of x is the “output.”
Again, we are forced to execute in order.
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

15. Why Does Order Matter?
Dependencies can affect whether we can execute a particular part of the program in parallel.
If we cannot execute that part of the program in parallel, then it’ll be SLOW.
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

16. Loop Dependency Example
if ((dst == src1) && (dst == src2)) {
for (index = 1; index < length; index++) {
dst[index] = dst[index-1] + dst[index]; }
else if (dst == src1) {
dst[index] = dst[index-1] + src2[index];
else if (dst == src2) {
dst[index] = src1[index-1] + dst[index];
else if (src1 == src2) {
dst[index = src1[index-1] + src1[index];
else {
dst[index] = src1[index-1] + src2[index];
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

17. Loop Dep Example (cont’d)
if ((dst == src1) && (dst == src2)) {
for (index = 1; index < length; index++) {
dst[index] = dst[index-1] + dst[index]; }
else if (dst == src1) {
dst[index] = dst[index-1] + src2[index];
else if (dst == src2) {
dst[index] = src1[index-1] + dst[index];
else if (src1 == src2) {
dst[index = src1[index-1] + src1[index];
else {
dst[index] = src1[index-1] + src2[index];
The various versions of the loop either:
do have loop carried dependencies, or
don’t have loop carried dependencies.
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

18. Loop Dependency Performance
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

19. Stupid Compiler Tricks

20. Stupid Compiler Tricks
Tricks Compilers Play
Scalar Optimizations
Loop Optimizations
Inlining
Tricks You Can Play with Compilers
Profiling
Hardware counters
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

21. Compiler Design
The people who design compilers have a lot of experience working with the languages commonly used in High Performance Computing:
Fortran: 45ish years
C: ish years
C++: ish years, plus C experience
So, they’ve come up with clever ways to make programs run faster.
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

22. Tricks Compilers Play

23. Scalar Optimizations Copy Propagation Constant Folding
Dead Code Removal
Strength Reduction
Common Subexpression Elimination
Variable Renaming
Loop Optimizations
Not every compiler does all of these, so it sometimes can be worth doing these by hand.
Much of this discussion is from [2] and [5].
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

24. Copy Propagation x = y z = 1 + x Before Has data dependency x = y
Compile
x = y
z = 1 + y
After
No data dependency
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

25. Constant Folding Before After add = 100 aug = 200 sum = add + aug
Notice that sum is actually the sum of two constants, so the compiler can precalculate it, eliminating the addition that otherwise would be performed at runtime.
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

26. Dead Code Removal Before After
var = 5
PRINT *, var
STOP
PRINT *, var * 2
var = 5
PRINT *, var
STOP
Since the last statement never executes, the compiler can eliminate it.
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

27. Strength Reduction Before After x = y ** 2.0 a = c / 2.0 x = y * y
Raising one value to the power of another, or dividing, is more expensive than multiplying. If the compiler can tell that the power is a small integer, or that the denominator is a constant, it’ll use multiplication instead.
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

28. Common Subexpression Elimination
Before
After
d = c * (a / b)
e = (a / b) * 2.0
adivb = a / b
d = c * adivb
e = adivb * 2.0
The subexpression (a / b) occurs in both assignment statements, so there’s no point in calculating it twice.
This is typically only worth doing if the common subexpression is expensive to calculate.
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

29. Variable Renaming Before After x = y * z q = r + x * 2 x = a + b
The original code has an output dependency, while the new code doesn’t – but the final value of x is still correct.
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

30. Loop Optimizations Hoisting Loop Invariant Code Unswitching
Iteration Peeling
Index Set Splitting
Loop Interchange
Unrolling
Loop Fusion
Loop Fission
Not every compiler does all of these, so it sometimes can be worth doing some of these by hand.
Much of this discussion is from [3] and [5].
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

31. Hoisting Loop Invariant Code
DO i = 1, n
a(i) = b(i) + c * d
e = g(n)
END DO
Code that doesn’t change inside the loop is called loop invariant. It doesn’t need to be calculated over and over.
Before
temp = c * d
DO i = 1, n
a(i) = b(i) + temp
END DO
e = g(n)
After
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

32. Unswitching The condition is j-independent. Before
DO i = 1, n
DO j = 2, n
IF (t(i) > 0) THEN
a(i,j) = a(i,j) * t(i) + b(j)
ELSE
a(i,j) = 0.0
END IF
END DO
Before
So, it can migrate outside the j loop.
After
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

33. Iteration Peeling Before
DO i = 1, n
IF ((i == 1) .OR. (i == n)) THEN
x(i) = y(i)
ELSE
x(i) = y(i + 1) + y(i – 1)
END IF
END DO
Before
We can eliminate the IF by peeling the weird iterations.
x(1) = y(1)
DO i = 2, n - 1
x(i) = y(i + 1) + y(i – 1)
END DO
x(n) = y(n)
After
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

34. Index Set Splitting Before After
DO i = 1, n
a(i) = b(i) + c(i)
IF (i > 10) THEN
d(i) = a(i) + b(i – 10)
END IF
END DO
DO i = 1, 10
DO i = 11, n
Before
After
Note that this is a generalization of peeling.
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

35. Loop Interchange After Before
DO j = 1, nj
DO i = 1, ni
a(i,j) = b(i,j)
END DO
DO i = 1, ni
DO j = 1, nj
a(i,j) = b(i,j)
END DO
Array elements a(i,j) and a(i+1,j) are near each other in memory, while a(i,j+1) may be far, so it makes sense to make the i loop be the inner loop. (This is reversed in C, C++ and Java.)
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

36. Unrolling DO i = 1, n a(i) = a(i)+b(i) Before END DO DO i = 1, n, 4
After
You generally shouldn’t unroll by hand.
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

37. Why Do Compilers Unroll?
We saw last time that a loop with a lot of operations gets better performance (up to some point), especially if there are lots of arithmetic operations but few main memory loads and stores.
Unrolling creates multiple operations that typically load from the same, or adjacent, cache lines.
So, an unrolled loop has more operations without increasing the memory accesses by much.
Also, unrolling decreases the number of comparisons on the loop counter variable, and the number of branches to the top of the loop.
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

38. Loop Fusion
DO i = 1, n
a(i) = b(i) + 1
END DO
c(i) = a(i) / 2
d(i) = 1 / c(i)
As with unrolling, this has fewer branches. It also has fewer total memory references.
Before
After
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

39. Loop Fission
DO i = 1, n
a(i) = b(i) + 1
c(i) = a(i) / 2
d(i) = 1 / c(i)
END DO !! i = 1, n
Fission reduces the cache load and the number of operations per iteration.
Before
After
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

40. To Fuse or to Fizz?
The question of when to perform fusion versus when to perform fission, like many many optimization questions, is highly dependent on the application, the platform and a lot of other issues that get very, very complicated.
Compilers don’t always make the right choices.
That’s why it’s important to examine the actual behavior of the executable.
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

41. Inlining
Before
After
DO i = 1, n
a(i) = func(i)
END DO …
REAL FUNCTION func (x)
func = x * 3
END FUNCTION func
DO i = 1, n
a(i) = i * 3
END DO
When a function or subroutine is inlined, its contents are transferred directly into the calling routine, eliminating the overhead of making the call.
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

42. Tricks You Can Play with Compilers

43. The Joy of Compiler Options
Every compiler has a different set of options that you can set.
Among these are options that control single processor optimization: superscalar, pipelining, vectorization, scalar optimizations, loop optimizations, inlining and so on.
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

44. Example Compile Lines IBM Regatta xlf90 –O –qmaxmem=-1 –qarch=auto
–qtune=auto –qcache=auto –qhot
Intel
ifort –O –tpp7 -xW
Portland Group f90
pgf90 -fastsse –Mdalign –Mvect=assoc
NAG f95
f95 –O4 –Ounsafe –ieee=nonstd
SGI Origin2000
f90 –Ofast –ipa
Sun UltraSPARC
f90 –fast
CrayJ90
f90 –O 3,aggress,pattern,recurrence
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

45. What Does the Compiler Do?
Example: NAG f95 compiler
f95 –O<level> source.f90
Possible levels are –O0, -O1, -O2, -O3, -O4:
-O0 No optimisation. …
-O1 Minimal quick optimisation.
-O2 Normal optimisation.
-O3 Further optimisation.
-O4 Maximal optimisation.[4]
The man page is pretty cryptic.
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

46. Arithmetic Operation Speeds
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

47. Optimization Performance
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

48. More Optimized Performance
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

49. Profiling

50. Profiling
Profiling means collecting data about how a program executes.
The two major kinds of profiling are:
Subroutine profiling
Hardware timing
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

51. Subroutine Profiling
Subroutine profiling means finding out how much time is spent in each routine.
The Rule: Typically, a program spends 90% of its runtime in 10% of the code.
Subroutine profiling tells you what parts of the program to spend time optimizing and what parts you can ignore.
Specifically, at regular intervals (e.g., every millisecond), the program takes note of what instruction it’s currently on.
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

52. Profiling Example On the IBM Regatta: xlf90 –O –g -pg …
The –g -pg options tell the compiler to set the executable up to collect profiling information.
Running the executable generates a file named gmon.out, which contains the profiling information.
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

53. Profiling Example (cont’d)
When the run has completed, a file named gmon.out has been generated.
Then:
gprof executable
produces a list of all of the routines and how much time was spent in each.
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

54. Profiling Result ... % cumulative self self total
time seconds seconds calls ms/call ms/call name
longwave_ [5]
mpdata3_ [8]
turb_ [9]
turb_scalar_ [10]
advect2_z_ [12]
cloud_ [11]
radiation_ [3]
smlr_ [7]
tke_full_ [13]
shear_prod_ [15]
rhs_ [16]
advect2_xy_ [17]
poisson_ [14]
long_wave_ [4]
advect_scalar_ [6]
buoy_ [19]
...
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

55. Hardware Timing
In addition to learning about which routines dominate in your program, you might also want to know how the hardware behaves; e.g., you might want to know how often you get a cache miss.
Many supercomputer CPUs have special hardware that measures such events, called event counters.
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

56. Hardware Timing Example
On SGI Origin2000:
perfex –x –a executable
This command produces a list of hardware counts.
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

57. Hardware Timing Results
Cycles
Decoded instructions
Decoded loads
Decoded stores
Grad floating point instructions
Primary data cache misses
Secondary data cache misses
. . .
This hardware counter profile shows that only 1% of memory accesses resulted in L2 cache misses, which is good, but that it only got 0.42 FLOPs per cycle, out of a peak of 2 FLOPs per cycle, which is bad.
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

58. Shared Memory Multithreading
Next Time
Part V:
Shared Memory Multithreading
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research

59. References
[1] Kevin Dowd and Charles Severance, High Performance Computing,
2nd ed. O’Reilly, 1998, p
[2] Ibid, p
[3] Ibid, p
[4] NAG f95 man page.
[5] Michael Wolfe, High Performance Compilers for Parallel Computing, Addison-Wesley Publishing Co., 1996.
[6] Kevin R. Wadleigh and Isom L. Crawford, Software Optimization for High Performance Computing, Prentice Hall PTR, 2000, pp
Supercomputing in Plain English: Compilers
OU Supercomputing Center for Education & Research